tl;dr — Week 33, 2025

Welcome to my latest experiment, a bi-weekly newsletter called Thib loved do read aka tl;dr

Thibault Maekelbergh

25 Aug 2025 • 5 min read

What is tl;dr, an introduction

Welcome to my latest experiment, a bi-weekly newsletter called Thib loved do read aka tl;dr

Anyone who really knows me knows I'm a lists guy. I make them for literally everything. And I'm also someone who can't help but share cool stuff I stumble across - a brilliant article, a new release of a piece of software I got excited about, a cool video I watched and learned a thing or two from, or just something that made me think "nice, that's clever."

So I've been thinking for a while now: why not combine these two things and share my lists?

I'll be hosting my own ~~curated list~~ newsletter that collects nice articles, product launches, release notes, random columns — basically all the stuff that I really liked reading and .

💡

What topics can you expect in tl;dr?
Well since I'm currently really into reading how AI blends with businesses, how it all works under the hood there's going be a lot of that content

Where I want to take this

I aim to make it a bit of a personal experiment as well, where I will try to use an AI assistant to help me write the actual newsletter

For a while I've been toying with this idea to make an LLM aware of my voice of tone and writing style by feeding it my earlier written content, so that afterwards I can simply feed it my Karakeep list, notes of the week or maybe even Safari tabs on my phone and see if it can get the newsletter drafted for me.

Ideally I can just make micro-edits and post my ✨ editorial notes ✨ & thoughts. I'm not quite sure if I'll like it but let's see, the pivot to writing these myself is easy enough to make 😺

_{I won't be using AI right off the bat and will write them myself first. If I change that it will be indicated clearly in the email.}

Here's what I liked reading

Simple AI mind tricks with big consequences

It lays out the basics of prompt injection but, more importantly, demonstrates once again that no company, big or small, is safe from what is essentially a very basic AI mind-trick.

Home Assistant keeps making home automation AI friendly

I like the steps HA is taking to make AI feel more embedded in the platform, especially by focusing on local AI solutions like Ollama in their examples. With the introduction of AI Tasks, I am eager to get my hands on it and start thinking about and creating possible tasks such as weather reports, notifications, and schedule briefings using my locally hosted models.

A new edge model from Google

Google introduced a new lightweight/edge model variant of Gemma 3. Don't expect it to be very good as an assistant or to answer questions correctly, but I'm looking forward to try it out in Home Assistant's AI tasks and see how it does, or finetune it on my own content 🤔. Given that it’s a 270-million-parameter model, it should run quickly enough locally on recent Mac hardware.

Mistral Document AI lands on Azure

I have long been a fan of Mistral. Releasing their Document AI offering on Azure will surely open doors for companies that may be hesitant to send data to Mistral's own cloud (La Plateforme) but are looking to leverage AI-enhanced document workflows.

AI and energy

Wendover Productions' videos are always a great source of learning new things for me. In this video, he explains how AI consumes energy from the grid and what that means for households. It's a very American problem (for now), but it provides a real understanding of what it means for our global energy infrastructure.

A very brief & clear warning on agentic AI

I've written a short bit on this exact video on LinkedIn before, but when paired with the first article in this newsletter, it becomes highly relevant again. I felt it was important to include it again here.

How digital time works

I knew what NTP was, but not how it worked. This short 8 min video breaks down the different layers of time syncing (Stratum levels) and explains how all of the digital clocks on our laptops, phones, or connected watches try to stay in sync with the correct time.